[CRIU] [PATCH v3 18/33] proto: Add ns_hookup_entry description

Kirill Tkhai ktkhai at virtuozzo.com
Mon Feb 20 00:39:48 PST 2017


On 17.02.2017 21:25, Andrei Vagin wrote:
> On Fri, Feb 17, 2017 at 11:05:14AM +0300, Kirill Tkhai wrote:
>>
>>
>> On 16.02.2017 22:59, Andrei Vagin wrote:
>>> On Thu, Feb 16, 2017 at 03:09:14PM +0300, Kirill Tkhai wrote:
>>>> New image format, generic for all namespaces.
>>>> Currently, it's for pid, net and user ns.
>>>>
>>>> v3: Move parent_id to pid and user ext
>>>>
>>>> Signed-off-by: Kirill Tkhai <ktkhai at virtuozzo.com>
>>>> ---
>>>>  criu/image-desc.c            |    1 +
>>>>  criu/include/image-desc.h    |    1 +
>>>>  criu/include/magic.h         |    1 +
>>>>  criu/include/protobuf-desc.h |    1 +
>>>>  criu/protobuf-desc.c         |    1 +
>>>>  images/ns.proto              |   19 +++++++++++++++++++
>>>>  lib/py/images/images.py      |    1 +
>>>>  7 files changed, 25 insertions(+)
>>>>
>>>> diff --git a/criu/image-desc.c b/criu/image-desc.c
>>>> index fffaf5a01..336a50e6d 100644
>>>> --- a/criu/image-desc.c
>>>> +++ b/criu/image-desc.c
>>>> @@ -99,6 +99,7 @@ struct cr_fd_desc_tmpl imgset_template[CR_FD_MAX] = {
>>>>  	FD_ENTRY(USERNS,	"userns-%d"),
>>>>  	FD_ENTRY(NETNF_CT,	"netns-ct-%d"),
>>>>  	FD_ENTRY(NETNF_EXP,	"netns-exp-%d"),
>>>> +	FD_ENTRY(NS_HOOKUP,	"ns-hookup"),
>>>>
>>>>  	[CR_FD_STATS] = {
>>>>  		.fmt	= "stats-%s",
>>>> diff --git a/criu/include/image-desc.h b/criu/include/image-desc.h
>>>> index 09d187d22..16d0183ff 100644
>>>> --- a/criu/include/image-desc.h
>>>> +++ b/criu/include/image-desc.h
>>>> @@ -107,6 +107,7 @@ enum {
>>>>  	CR_FD_EVENTPOLL_TFD,
>>>>
>>>>  	CR_FD_AUTOFS,
>>>> +	CR_FD_NS_HOOKUP,
>>>>
>>>>  	CR_FD_MAX
>>>>  };
>>>> diff --git a/criu/include/magic.h b/criu/include/magic.h
>>>> index deb54b1d0..34e766fa3 100644
>>>> --- a/criu/include/magic.h
>>>> +++ b/criu/include/magic.h
>>>> @@ -93,6 +93,7 @@
>>>>  #define SECCOMP_MAGIC		0x64413049 /* Kostomuksha */
>>>>  #define BINFMT_MISC_MAGIC	0x67343323 /* Apatity */
>>>>  #define AUTOFS_MAGIC		0x49353943 /* Sochi */
>>>> +#define NS_HOOKUP_MAGIC		0x55523343 /* Dnepropetrovsk */
>>>>
>>>>  #define IFADDR_MAGIC		RAW_IMAGE_MAGIC
>>>>  #define ROUTE_MAGIC		RAW_IMAGE_MAGIC
>>>> diff --git a/criu/include/protobuf-desc.h b/criu/include/protobuf-desc.h
>>>> index 6c76b494f..e45700b3d 100644
>>>> --- a/criu/include/protobuf-desc.h
>>>> +++ b/criu/include/protobuf-desc.h
>>>> @@ -70,6 +70,7 @@ enum {
>>>>  	PB_SK_QUEUES,
>>>>  	PB_IPCNS_MSG,
>>>>  	PB_IPCNS_MSG_ENT,
>>>> +	PB_NS_HOOKUP,
>>>>
>>>>  	PB_MAX,
>>>>  };
>>>> diff --git a/criu/protobuf-desc.c b/criu/protobuf-desc.c
>>>> index 05f3eb386..f3361ff5f 100644
>>>> --- a/criu/protobuf-desc.c
>>>> +++ b/criu/protobuf-desc.c
>>>> @@ -97,6 +97,7 @@ void cr_pb_init(void)
>>>>  	CR_PB_DESC(REMAP_FPATH,		RemapFilePath,	remap_file_path);
>>>>  	CR_PB_DESC(NETDEV,		NetDevice,	net_device);
>>>>  	CR_PB_MDESC_INIT(cr_pb_descs[PB_PAGEMAP_HEAD],	PagemapHead,	pagemap_head);
>>>> +	CR_PB_DESC(NS_HOOKUP,		NsHookup,	ns_hookup);
>>>>
>>>>  #include "protobuf-desc-gen.h"
>>>>  }
>>>> diff --git a/images/ns.proto b/images/ns.proto
>>>> index 7f7b09094..01cb032bf 100644
>>>> --- a/images/ns.proto
>>>> +++ b/images/ns.proto
>>>> @@ -16,4 +16,23 @@ message uid_gid_extent {
>>>>  message userns_entry {
>>>>  	repeated uid_gid_extent uid_map		= 1;
>>>>  	repeated uid_gid_extent gid_map 	= 2;
>>>> +	optional uint32		parent_id	= 3;
>>>
>>> I've ignored my comment to the previous version:
>>>
>>> """
>>> I think we need to move parent_id to pid_ext, because it will be used
>>> only for pid namespaces.
>>>
>>> And I think userns_id has to be here, because it is required for all
>>> namespaces. In this case we will not need to create netns_entry2.
>>>
>>> What do you think about this?
>>> """
>>
>> I didn't ignore. Haven't you seen my reply on it?
>>
>> "Hm, but user_ns has no link to a user_ns, so it's not generic."
> 
> parent and user_ns for user_ns have the same meaning.

I'm not agree. It's not a common truth, it depends on a treatment.
For net_ns user_ns is a propertry, while for user_ns it's a member
of a hierarhy. Also, pid_ns has both parent and user_ns pointers,
and in the way you suggest it would be asymmetric to user_ns.

Pasha said, the main rule is to try to minimize number of files
in dump. This way, some days netns_entry should be merged to netns_entry2,
so this field will exist anyway.


If you want to simplify ns file writing/reading, we may make both
of .user_ns and .parent fields common for all namespaces, but not
only one of them.

>>
>>>> +}
>>>> +
>>>> +message pidns_entry {
>>>> +	required uint32	userns_id		= 1;
>>>> +	optional uint32		parent_id	= 2;
>>>> +}
>>>> +
>>>> +message netns_entry2 {
>>>> +	required uint32	userns_id		= 1;
>>>> +}
>>>> +
>>>> +message ns_hookup_entry {
>>>> +	required uint32		id		= 1;
>>>> +	required uint32		ns_cflag	= 2;
>>>> +
>>>> +	optional userns_entry	user_ext	= 3;
>>>> +	optional pidns_entry	pid_ext		= 4;
>>>> +	optional netns_entry2	net_ext		= 5;
>>>>  }
>>>> diff --git a/lib/py/images/images.py b/lib/py/images/images.py
>>>> index c593a3b85..69f813c40 100644
>>>> --- a/lib/py/images/images.py
>>>> +++ b/lib/py/images/images.py
>>>> @@ -454,6 +454,7 @@ handlers = {
>>>>  	'USERNS'		: entry_handler(userns_entry),
>>>>  	'SECCOMP'		: entry_handler(seccomp_entry),
>>>>  	'AUTOFS'		: entry_handler(autofs_entry),
>>>> +	'NS_HOOKUP'	        : entry_handler(ns_hookup_entry),
>>>>  	}
>>>>
>>>>  def __rhandler(f):
>>>>
>>>> _______________________________________________
>>>> CRIU mailing list
>>>> CRIU at openvz.org
>>>> https://lists.openvz.org/mailman/listinfo/criu


More information about the CRIU mailing list